Complete OpenAI to AWS Bedrock conversion

[yuzisun]

• Complete OpenAI Chat completion and AWSBedrock Converse type definition
• Additional conversion for InferenceConfiguration and ToolConfig
• Error handling and convert to OpenAI error
• Convert to table testing and add more tests

[mathetake]

@yuzisun the e2e tests are failing https://github.com/envoyproxy/ai-gateway/actions/runs/12624576475/job/35175024164?pr=64 due to some parsing errors

ai-gateway/tests/extproc/extproc_test.go

Lines 73 to 101 in 1390554

	t.Run("health-checking", func(t *testing.T) {
	client := openai.NewClient(option.WithBaseURL("http://localhost:1062/v1/"))
	for _, tc := range []struct {
	testCaseName,
	modelName string
	}{
	{testCaseName: "openai", modelName: "gpt-4o-mini"}, // This will go to "openai"
	{testCaseName: "aws-bedrock", modelName: "us.meta.llama3-2-1b-instruct-v1:0"}, // This will go to "aws-bedrock".
	} {
	t.Run(tc.modelName, func(t *testing.T) {
	require.Eventually(t, func() bool {
	chatCompletion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
	Messages: openai.F([]openai.ChatCompletionMessageParamUnion{
	openai.UserMessage("Say this is a test"),
	}),
	Model: openai.F(tc.modelName),
	})
	if err != nil {
	t.Logf("error: %v", err)
	return false
	}
	for _, choice := range chatCompletion.Choices {
	t.Logf("choice: %s", choice.Message.Content)
	}
	return true
	}, 10*time.Second, 1*time.Second)
	})
	}
	})

[yuzisun]

@yuzisun the e2e tests are failing https://github.com/envoyproxy/ai-gateway/actions/runs/12624576475/job/35175024164?pr=64 due to some parsing errors

ai-gateway/tests/extproc/extproc_test.go

Lines 73 to 101 in 1390554

	t.Run("health-checking", func(t *testing.T) {
	client := openai.NewClient(option.WithBaseURL("http://localhost:1062/v1/"))
	for _, tc := range []struct {
	testCaseName,
	modelName string
	}{
	{testCaseName: "openai", modelName: "gpt-4o-mini"}, // This will go to "openai"
	{testCaseName: "aws-bedrock", modelName: "us.meta.llama3-2-1b-instruct-v1:0"}, // This will go to "aws-bedrock".
	} {
	t.Run(tc.modelName, func(t *testing.T) {
	require.Eventually(t, func() bool {
	chatCompletion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
	Messages: openai.F([]openai.ChatCompletionMessageParamUnion{
	openai.UserMessage("Say this is a test"),
	}),
	Model: openai.F(tc.modelName),
	})
	if err != nil {
	t.Logf("error: %v", err)
	return false
	}
	for _, choice := range chatCompletion.Choices {
	t.Logf("choice: %s", choice.Message.Content)
	}
	return true
	}, 10*time.Second, 1*time.Second)
	})
	}
	})

Ah looks like you are using the new type ChatCompletionNewParams instead of ChatCompletion in the test.

[mathetake]

yeah - so we need basically the manual type checking logic to support unions on .contents to be compatible with openai i guess. the ChatCompletionNewParams can actually be accepted by them and that use is the example from go-sdk https://github.com/openai/openai-go?tab=readme-ov-file#usage

[mathetake]

you can ignore Tests using Secrets / External Processor (pull_request) (fixed in #68) and Tests using Secrets / External Processor (pull_request_target) is what matters. It's still failing but it seems almost there?

[yuzisun]

you can ignore Tests using Secrets / External Processor (pull_request) (fixed in #68) and Tests using Secrets / External Processor (pull_request_target) is what matters. It's still failing but it seems almost there?

@mathetake The new schema is quite complex which took me a little more time, unfortunately we have to manually craft the types as openai-go did not expose the unmarshall APIs. I need to run the e2e test again, how do I trigger it ?

[mathetake]

@yuzisun thank you for the excellent job here!!